Model Context Protocol (MCP) Release Date & History 2026

When Was MCP Released?

Model Context Protocol (MCP) was officially released by Anthropic on November 25, 2024. The open-source specification and initial SDK dropped alongside Claude Desktop support, marking the first time a major AI lab published a standardized protocol for connecting LLMs to external tools and data sources.

The initial release included:

The MCP specification (JSON-RPC 2.0 based)
Python SDK and TypeScript SDK
A set of reference servers (filesystem, GitHub, Slack, PostgreSQL, SQLite)
Claude Desktop as the first MCP host

MCP Full Timeline: November 2024 to 2026

November 25, 2024

Initial Public Release by Anthropic

Anthropic open-sourced MCP under the MIT license. The announcement described MCP as "a new standard for connecting AI assistants to the systems where data lives." Claude Desktop became the first MCP-compatible host application.

December 2024

Early Ecosystem Adoption

Within weeks, the developer community started building custom MCP servers. Early adopters included Block (formerly Square), Apollo, and several open-source contributors. The MCP GitHub repository crossed 5,000 stars within 30 days of launch.

January 2025

Claude.ai Web Support + Expanded SDKs

Anthropic extended MCP support to Claude.ai in the browser. Java and Kotlin SDKs were added alongside the original Python and TypeScript packages, broadening the server development audience.

February 2025

Third-Party Host Integrations Begin

Zed (the code editor), Replit, Codeium, and Sourcegraph announced MCP support, demonstrating that the protocol was not Claude-exclusive. This was the first major signal that MCP could become an industry standard rather than an Anthropic proprietary format.

March 2025

OpenAI Adopts MCP

OpenAI announced support for MCP in the Agents SDK and ChatGPT desktop app — a watershed moment confirming industry-wide adoption. With both Anthropic and OpenAI backing the standard, MCP rapidly became the de facto protocol for AI tool integrations.

April 2025

Remote MCP Servers + Auth Spec

The MCP spec was updated to support remote servers over HTTP/SSE (in addition to local stdio), along with OAuth 2.0 authentication. This unlocked enterprise and SaaS use cases where servers could not run locally. Major SaaS vendors began announcing hosted MCP endpoints.

May 2025

1,000+ Community Servers

The community-curated MCP server list surpassed 1,000 entries. Google DeepMind, Microsoft, and Meta all publicly acknowledged MCP compatibility plans. mcp.so and mcpservers.org directories launched to help developers discover servers.

Q3–Q4 2025

MCP Goes Enterprise

Enterprise-focused features landed: sampling (servers can call back to LLMs), roots (servers declare filesystem scope), and progress notifications for long-running operations. Microsoft Copilot Studio, Salesforce, and ServiceNow integrated MCP into their enterprise AI platforms.

Q1 2026

MCP 1.0 Stable Specification

The MCP working group (including contributors from Anthropic, OpenAI, Google, and Microsoft) released the 1.0 stable specification. This formalized the transport layer (stdio, HTTP+SSE, WebSocket), capability negotiation, and error handling conventions that had emerged from community practice.

2026 (Current)

2,500+ Servers, All Major AI Platforms

As of 2026, MCP is supported by all major AI assistant platforms. The ecosystem includes 2,500+ community and vendor-maintained servers. MCP Inspector, FastMCP, and high-level orchestration frameworks (LangChain MCP adapter, LlamaIndex MCP tools) have made development accessible to non-specialist developers.

Why Did MCP Take Off So Quickly?

Several factors drove rapid adoption:

Open source from day one — MIT license meant no vendor lock-in concerns
Simple protocol — JSON-RPC 2.0 over stdio or HTTP is easy to implement in any language
Anthropic credibility — Launching alongside Claude gave it immediate real-world usage
OpenAI endorsement — March 2025 adoption removed the "Anthropic-only" stigma
Right timing — The market needed a standard as AI agent use cases exploded in 2024-2025

MCP vs Pre-MCP: What Changed?

Before MCP	After MCP
Each LLM app needed custom integrations for every tool	One MCP server works with any MCP-compatible host
Function calling was model-specific (OpenAI format vs others)	Unified tool/resource/prompt primitives across models
No standard for resource access (files, DBs, APIs)	Resources primitive provides structured data access
Integrations were brittle, hard to maintain	Versioned spec with backward compatibility guarantees
Enterprise security was an afterthought	Built-in OAuth 2.0, roots scoping, sampling controls

Key MCP Concepts (Quick Reference)

Tools — Functions the LLM can call (like function calling, but standardized)
Resources — Structured data the LLM can read (files, DB rows, API results)
Prompts — Reusable prompt templates with arguments
Sampling — Servers can request LLM completions (agentic loops)
Roots — Filesystem scope declarations for security
Transports — stdio (local), HTTP+SSE (remote), WebSocket

Top MCP Servers to Know in 2026

The most widely used official and community MCP servers:

filesystem — Read/write local files with root scoping
github — Repos, PRs, issues, code search
postgresql — Database queries with schema introspection
brave-search — Web search integration
puppeteer / playwright — Browser automation
slack — Channel messages and DMs
google-drive — Docs, Sheets, files
memory — Persistent key-value memory for agents

Browse the full directory of 2,500+ MCP tools at AgDex.ai MCP Tools.